Skip to content

Are your APIs ready to be called by AI agents?

Mykhailo PoliarushUpdated 5 min read

In 2026, API integration is the bottleneck between companies and their AI plans. The average enterprise runs 957 applications and only 27% of them are integrated, according to the Salesforce 2026 Connectivity Benchmark. At the same time, APIs have a new kind of caller, the AI agent, and most APIs and security programs were not built for it.

Why has integration become the bottleneck for AI?

Agents are only as useful as the systems they can reach. In the Salesforce survey of 1,050 enterprise IT leaders (fielded October to November 2025; Salesforce owns MuleSoft and sells integration software), 96% said AI agent success depends on data integration across all systems, and 94% said it will require a more API-driven architecture. The app count grew from 897 to 957 in a year, so the integration gap is widening, not closing.

Integration also eats engineering time. In the 2025 edition of the same report, IT teams spent 39% of their time designing, building and testing custom integrations, and 95% of respondents struggled to integrate data across systems. Nearly half (49%) of organizations cite cross-application data governance as a top integration challenge, per the Salesforce 2026 survey, and 86% of IT leaders worry that agents will add more complexity than value without proper integration. Governance lags behind: an estimated 27% of APIs are ungoverned on average, per the 2026 report.

Who is calling your APIs now?

More and more often, software that acts on its own. APIs already carry most of the web's dynamic traffic: 57%, according to Cloudflare's API security report, based on its network data from October 2022 to August 2023. They also make money: 65% of organizations generate revenue from their API programs, according to Postman's 2025 State of the API report, a survey of more than 5,700 developers, architects and executives by an API platform vendor.

Design has not caught up with the new consumer. In the Postman report, 89% of developers use AI, but only 24% design APIs with AI agents in mind, and 60% still design primarily for humans. That matters because an agent cannot ask a colleague what an error code means. It needs predictable schemas, typed errors and documentation that says when and why to call an endpoint.

The API-first shift helps. Postman found that 82% of organizations have adopted some level of API-first development and 25% are fully API-first, which Postman reports as a 12% increase from 2024.

Where does MCP fit, and who is using it?

The Model Context Protocol (MCP) is becoming the standard way to give agents access to tools and APIs. It had more than 10,000 active public servers when Anthropic donated it to the Linux Foundation's Agentic AI Foundation in December 2025 (Anthropic's own figures; see our state of AI engineering report for the wider picture).

Day-to-day use is still early. In Postman's survey, only 10% of developers used MCP regularly and another 24% planned to explore it, while 31% were not familiar with it.

How developers use the Model Context Protocol (MCP)

Not familiar with MCP
31%
Have not used it, plan to explore
24%
Used it occasionally for experiments or side projects
19%
Use it regularly in daily work
10%
Evaluated it, chose not to implement
7%
Used it in the past, no longer
5%
Have not used it, do not intend to
4%

Source: Postman, 2025 State of the API Report, vendor survey (n=5,700+, 2025)

The servers that exist often handle credentials badly. Astrix Security, a vendor of identity security tools, analyzed over 5,200 open-source MCP server implementations: 88% require credentials, 53% rely on long-lived static secrets such as API keys and personal access tokens, and only 8.5% use OAuth. Astrix used an LLM to classify README files, so treat the split as an estimate, but the direction is clear. An MCP server with a static admin token gives every agent that connects to it the same broad access.

What are the main API and integration challenges in 2026?

  1. App sprawl. With 957 applications and only 27% integrated in the average enterprise (Salesforce), most integration work is still ahead.
  2. APIs nobody knows about. Only 19% of security professionals are very confident in the accuracy of their API inventory, according to the Salt Security H2 2025 report (n=386, a vendor survey). Cloudflare found nearly 31% more REST endpoints through machine learning discovery than customers had declared.
  3. Security incidents and delays. In the Salt survey, 33% had an API security incident in the past year, 50% delayed a new application rollout over API security concerns, and 80% lacked continuous, real-time API monitoring.
  4. Agent-specific risk. The top concern in Postman's survey, cited by 51%, is unauthorized or excessive API calls from AI agents, followed by AI accessing sensitive data (49%) and AI leaking API credentials (46%).
  5. Collaboration. In the same Postman report, 93% of API teams face collaboration blockers, and the most common ones are inconsistent or outdated documentation and APIs that teams cannot find.

API security findings, share of organizations (separate questions)

Lack continuous, real-time API monitoring
80%
Delayed an app rollout over API security concerns
50%
Had an API security incident in the past year
33%
Very confident in their API inventory
19%

Source: Salt Security, H2 2025 State of API Security Report, vendor survey (n=386, Oct 2025)

What does good API and integration practice look like?

Treat APIs as long-lived products with owners, contracts and tests. In practice:

  • Know what you have. Keep one catalog of internal and external APIs with an owner for each. Compare it with what your gateway and logs actually see.
  • Contract first. Write the OpenAPI (or AsyncAPI) spec before the code, review it, and generate docs and clients from it.
  • Test the contract on every change. Contract and integration tests in CI catch breaking changes before a partner or an agent does.
  • Design for machine callers. Consistent schemas, typed errors, explicit rate limits and clear descriptions of side effects. If an agent can read the spec and call the endpoint correctly, so can a new developer.
  • Give agents their own identity. Postman recommends identifying agent traffic, scoping keys to least privilege and monitoring for spikes in 401 and 403 errors. Prefer OAuth with narrow scopes over shared static keys, and expose only the tools you intend through MCP.
  • Monitor at runtime. Track error rates, latency and unusual call patterns per consumer, and alert on them.

To start, pick the integration that hurts most (the one that breaks often or blocks an AI use case), document its contract, put tests around it and fix its authentication. Then repeat for the next one.

Want a second opinion on your APIs and integrations?

SDCLabs has built and tested software since 2011. We are the founders of the open-source testing frameworks Codeception and CodeceptJS. Senior engineers run AI-assisted workflows, and automated tests verify every change before it ships. See our API and integration services or contact us for a first response within 2 hours on business days.

If agents are the reason you are looking at your APIs, see also our agentic process transformation service and AI agents work.

Sources checked on 2026-11-17.

Frequently asked questions

How many enterprise applications are integrated in 2026?

About a quarter. The average enterprise runs 957 applications and only 27% of them are integrated, according to the Salesforce 2026 Connectivity Benchmark, a vendor survey of 1,050 IT leaders.

Are APIs being designed for AI agents?

Rarely. In Postman's 2025 State of the API report, 89% of developers use AI but only 24% design APIs with AI agents in mind, and 60% still design primarily for humans.

How widely is MCP used by developers?

It is still early. Only 10% of developers in Postman's survey used MCP regularly, 24% planned to explore it and 31% were not familiar with it. An Astrix analysis of over 5,200 MCP servers found only 8.5% use OAuth.

How common are API security incidents?

In the Salt Security H2 2025 report, 33% of organizations had an API security incident in the past year and 50% delayed an application rollout over API security concerns. It is a vendor survey of 386 security professionals.

Related posts

Get an estimate for your product.